Mixed-frequency fusion grey panel model for spatiotemporal prediction of photovoltaic power generation

Accurate prediction of photovoltaic power generation (PPG) is vital for renewable energy stability, economic viability, and sustainable development. Existing energy prediction models rely on data sampled at single-frequency or single-frequency-multiple. To effectively address the spatiotemporal prediction challenges in PPG caused by varying sampling frequency differences (asynchronous mixed-frequency), this study first proposes a novel mixed-frequency fusion grey panel model. To improve accuracy, a two-stage parameter estimation combining quasi-maximum likelihood estimation with a meta-heuristic algorithm is developed, with unbiasedness, consistency, and efficiency validated through mathematical analysis and Monte Carlo simulations. Finally, using asynchronous mixed-frequency panel datasets from photovoltaic users in China, the new model is compared and empirically analyzed against eight benchmark models. Comparative results demonstrate that the new model exhibits significant advantages in prediction performance, stability, and generalization capability. It can directly utilize asynchronous high-frequency meteorological indicators, like weekly, ten-day, and monthly irradiation, wind speed, and precipitation to predict low-frequency PPG. Empirical results indicate that irradiation changes can rapidly affect PPG, while the impact of wind speed takes longer to manifest. Additionally, the spatial dependence of PPG is relatively limited, but historical cumulative effects significantly suppress the output. Furthermore, the future monthly PPG overall exhibits a seasonal downward trend.